/**
* Generate a stream of signatures/checksums that describe a buffer.
*
* Generate approximately one checksum every @p n bytes.
*
* @return Newly-allocated sum_struct
**/
static struct sum_struct *generate_sums(struct map_struct *buf, OFF_T len,
int n)
{
int i;
struct sum_struct *s;
int count;
int block_len = n;
int remainder = (len % block_len);
OFF_T offset = 0;
count = (len + (block_len - 1)) / block_len;
s = (struct sum_struct *) malloc(sizeof(*s));
if (!s)
out_of_memory("generate_sums");
s->count = count;
s->remainder = remainder;
s->n = n;
s->flength = len;
if (count == 0) {
s->sums = NULL;
return s;
}
if (verbose > 3) {
rprintf(FINFO, "count=%ld rem=%ld n=%ld flength=%.0f\n",
(long) s->count, (long) s->remainder,
(long) s->n, (double) s->flength);
}
s->sums = (struct sum_buf *) malloc(sizeof(s->sums[0]) * s->count);
if (!s->sums)
out_of_memory("generate_sums");
for (i = 0; i < count; i++) {
int n1 = MIN(len, n);
char *map = map_ptr(buf, offset, n1);
s->sums[i].sum1 = get_checksum1(map, n1);
get_checksum2(map, n1, s->sums[i].sum2);
s->sums[i].offset = offset;
s->sums[i].len = n1;
s->sums[i].i = i;
if (verbose > 3)
rprintf(FINFO,
"chunk[%d] offset=%.0f len=%d sum1=%08x\n",
i, (double) s->sums[i].offset,
s->sums[i].len, s->sums[i].sum1);
len -= n1;
offset += n1;
}
return s;
}
static void hash_search(int f,struct sum_struct *s,
struct map_struct *buf, OFF_T len)
{
OFF_T offset, aligned_offset, end;
int32 k, want_i, aligned_i, backup;
char sum2[SUM_LENGTH];
uint32 s1, s2, sum;
int more;
schar *map;
/* want_i is used to encourage adjacent matches, allowing the RLL
* coding of the output to work more efficiently. */
want_i = 0;
if (DEBUG_GTE(DELTASUM, 2)) {
rprintf(FINFO, "hash search b=%ld len=%s\n",
(long)s->blength, big_num(len));
}
k = (int32)MIN(len, (OFF_T)s->blength);
map = (schar *)map_ptr(buf, 0, k);
sum = get_checksum1((char *)map, k);
s1 = sum & 0xFFFF;
s2 = sum >> 16;
if (DEBUG_GTE(DELTASUM, 3))
rprintf(FINFO, "sum=%.8x k=%ld\n", sum, (long)k);
offset = aligned_offset = aligned_i = 0;
end = len + 1 - s->sums[s->count-1].len;
if (DEBUG_GTE(DELTASUM, 3)) {
rprintf(FINFO, "hash search s->blength=%ld len=%s count=%s\n",
(long)s->blength, big_num(len), big_num(s->count));
}
do {
int done_csum2 = 0;
uint32 hash_entry;
int32 i, *prev;
if (DEBUG_GTE(DELTASUM, 4)) {
rprintf(FINFO, "offset=%s sum=%04x%04x\n",
big_num(offset), s2 & 0xFFFF, s1 & 0xFFFF);
}
if (tablesize == TRADITIONAL_TABLESIZE) {
hash_entry = SUM2HASH2(s1,s2);
if ((i = hash_table[hash_entry]) < 0)
goto null_hash;
sum = (s1 & 0xffff) | (s2 << 16);
} else {
sum = (s1 & 0xffff) | (s2 << 16);
hash_entry = BIG_SUM2HASH(sum);
if ((i = hash_table[hash_entry]) < 0)
goto null_hash;
}
prev = &hash_table[hash_entry];
hash_hits++;
do {
int32 l;
/* When updating in-place, the chunk's offset must be
* either >= our offset or identical data at that offset.
* Remove any bypassed entries that we can never use. */
if (updating_basis_file && s->sums[i].offset < offset
&& !(s->sums[i].flags & SUMFLG_SAME_OFFSET)) {
*prev = s->sums[i].chain;
continue;
}
prev = &s->sums[i].chain;
if (sum != s->sums[i].sum1)
continue;
/* also make sure the two blocks are the same length */
l = (int32)MIN((OFF_T)s->blength, len-offset);
if (l != s->sums[i].len)
continue;
if (DEBUG_GTE(DELTASUM, 3)) {
rprintf(FINFO,
"potential match at %s i=%ld sum=%08x\n",
big_num(offset), (long)i, sum);
}
if (!done_csum2) {
map = (schar *)map_ptr(buf,offset,l);
get_checksum2((char *)map,l,sum2);
done_csum2 = 1;
}
if (memcmp(sum2,s->sums[i].sum2,s->s2length) != 0) {
false_alarms++;
continue;
}
/* When updating in-place, the best possible match is
* one with an identical offset, so we prefer that over
* the adjacent want_i optimization. */
if (updating_basis_file) {
/* All the generator's chunks start at blength boundaries. */
while (aligned_offset < offset) {
aligned_offset += s->blength;
aligned_i++;
}
if ((offset == aligned_offset
|| (sum == 0 && l == s->blength && aligned_offset + l <= len))
&& aligned_i < s->count) {
if (i != aligned_i) {
if (sum != s->sums[aligned_i].sum1
|| l != s->sums[aligned_i].len
|| memcmp(sum2, s->sums[aligned_i].sum2, s->s2length) != 0)
goto check_want_i;
i = aligned_i;
}
if (offset != aligned_offset) {
/* We've matched some zeros in a spot that is also zeros
* further along in the basis file, if we find zeros ahead
* in the sender's file, we'll output enough literal data
* to re-align with the basis file, and get back to seeking
* instead of writing. */
backup = (int32)(aligned_offset - last_match);
if (backup < 0)
backup = 0;
map = (schar *)map_ptr(buf, aligned_offset - backup, l + backup)
+ backup;
sum = get_checksum1((char *)map, l);
if (sum != s->sums[i].sum1)
goto check_want_i;
get_checksum2((char *)map, l, sum2);
if (memcmp(sum2, s->sums[i].sum2, s->s2length) != 0)
goto check_want_i;
/* OK, we have a re-alignment match. Bump the offset
* forward to the new match point. */
offset = aligned_offset;
}
/* This identical chunk is in the same spot in the old and new file. */
s->sums[i].flags |= SUMFLG_SAME_OFFSET;
want_i = i;
}
}
check_want_i:
/* we've found a match, but now check to see
* if want_i can hint at a better match. */
if (i != want_i && want_i < s->count
&& (!updating_basis_file || s->sums[want_i].offset >= offset
|| s->sums[want_i].flags & SUMFLG_SAME_OFFSET)
&& sum == s->sums[want_i].sum1
&& memcmp(sum2, s->sums[want_i].sum2, s->s2length) == 0) {
/* we've found an adjacent match - the RLL coder
* will be happy */
i = want_i;
}
want_i = i + 1;
matched(f,s,buf,offset,i);
offset += s->sums[i].len - 1;
k = (int32)MIN((OFF_T)s->blength, len-offset);
map = (schar *)map_ptr(buf, offset, k);
sum = get_checksum1((char *)map, k);
s1 = sum & 0xFFFF;
s2 = sum >> 16;
matches++;
break;
} while ((i = s->sums[i].chain) >= 0);
null_hash:
backup = (int32)(offset - last_match);
/* We sometimes read 1 byte prior to last_match... */
if (backup < 0)
backup = 0;
/* Trim off the first byte from the checksum */
more = offset + k < len;
map = (schar *)map_ptr(buf, offset - backup, k + more + backup)
+ backup;
s1 -= map[0] + CHAR_OFFSET;
s2 -= k * (map[0]+CHAR_OFFSET);
/* Add on the next byte (if there is one) to the checksum */
if (more) {
s1 += map[k] + CHAR_OFFSET;
s2 += s1;
} else
--k;
/* By matching early we avoid re-reading the
data 3 times in the case where a token
match comes a long way after last
match. The 3 reads are caused by the
running match, the checksum update and the
literal send. */
if (backup >= s->blength+CHUNK_SIZE && end-offset > CHUNK_SIZE)
matched(f, s, buf, offset - s->blength, -2);
} while (++offset < end);
matched(f, s, buf, len, -1);
map_ptr(buf, len-1, 1);
}
static void hash_search(int f,struct sum_struct *s,
struct map_struct *buf, OFF_T len)
{
OFF_T offset, end;
int32 k, want_i, backup;
char sum2[SUM_LENGTH];
uint32 s1, s2, sum;
int more;
schar *map;
/* want_i is used to encourage adjacent matches, allowing the RLL
* coding of the output to work more efficiently. */
want_i = 0;
if (verbose > 2) {
rprintf(FINFO, "hash search b=%ld len=%.0f\n",
(long)s->blength, (double)len);
}
k = (int32)MIN(len, (OFF_T)s->blength);
map = (schar *)map_ptr(buf, 0, k);
sum = get_checksum1((char *)map, k);
s1 = sum & 0xFFFF;
s2 = sum >> 16;
if (verbose > 3)
rprintf(FINFO, "sum=%.8x k=%ld\n", sum, (long)k);
offset = 0;
end = len + 1 - s->sums[s->count-1].len;
if (verbose > 3) {
rprintf(FINFO, "hash search s->blength=%ld len=%.0f count=%.0f\n",
(long)s->blength, (double)len, (double)s->count);
}
do {
int done_csum2 = 0;
int32 i;
if (verbose > 4) {
rprintf(FINFO, "offset=%.0f sum=%04x%04x\n",
(double)offset, s2 & 0xFFFF, s1 & 0xFFFF);
}
i = hash_table[SUM2HASH2(s1,s2)];
if (i < 0)
goto null_hash;
sum = (s1 & 0xffff) | (s2 << 16);
hash_hits++;
do {
int32 l;
if (sum != s->sums[i].sum1)
continue;
/* also make sure the two blocks are the same length */
l = (int32)MIN((OFF_T)s->blength, len-offset);
if (l != s->sums[i].len)
continue;
/* in-place: ensure chunk's offset is either >= our
* offset or that the data didn't move. */
if (updating_basis_file && s->sums[i].offset < offset
&& !(s->sums[i].flags & SUMFLG_SAME_OFFSET))
continue;
if (verbose > 3) {
rprintf(FINFO,
"potential match at %.0f i=%ld sum=%08x\n",
(double)offset, (long)i, sum);
}
if (!done_csum2) {
map = (schar *)map_ptr(buf,offset,l);
get_checksum2((char *)map,l,sum2);
done_csum2 = 1;
}
if (memcmp(sum2,s->sums[i].sum2,s->s2length) != 0) {
false_alarms++;
continue;
}
/* When updating in-place, the best possible match is
* one with an identical offset, so we prefer that over
* the following want_i optimization. */
if (updating_basis_file) {
int32 i2;
for (i2 = i; i2 >= 0; i2 = s->sums[i2].chain) {
if (s->sums[i2].offset != offset)
continue;
if (i2 != i) {
if (sum != s->sums[i2].sum1)
break;
if (memcmp(sum2, s->sums[i2].sum2,
s->s2length) != 0)
break;
i = i2;
}
/* This chunk was at the same offset on
* both the sender and the receiver. */
s->sums[i].flags |= SUMFLG_SAME_OFFSET;
goto set_want_i;
}
}
/* we've found a match, but now check to see
* if want_i can hint at a better match. */
if (i != want_i && want_i < s->count
&& (!updating_basis_file || s->sums[want_i].offset >= offset
|| s->sums[want_i].flags & SUMFLG_SAME_OFFSET)
&& sum == s->sums[want_i].sum1
&& memcmp(sum2, s->sums[want_i].sum2, s->s2length) == 0) {
/* we've found an adjacent match - the RLL coder
* will be happy */
i = want_i;
}
set_want_i:
want_i = i + 1;
matched(f,s,buf,offset,i);
offset += s->sums[i].len - 1;
k = (int32)MIN((OFF_T)s->blength, len-offset);
map = (schar *)map_ptr(buf, offset, k);
sum = get_checksum1((char *)map, k);
s1 = sum & 0xFFFF;
s2 = sum >> 16;
matches++;
break;
} while ((i = s->sums[i].chain) >= 0);
null_hash:
backup = (int32)(offset - last_match);
/* We sometimes read 1 byte prior to last_match... */
if (backup < 0)
backup = 0;
/* Trim off the first byte from the checksum */
more = offset + k < len;
map = (schar *)map_ptr(buf, offset - backup, k + more + backup)
+ backup;
s1 -= map[0] + CHAR_OFFSET;
s2 -= k * (map[0]+CHAR_OFFSET);
/* Add on the next byte (if there is one) to the checksum */
if (more) {
s1 += map[k] + CHAR_OFFSET;
s2 += s1;
} else
--k;
/* By matching early we avoid re-reading the
data 3 times in the case where a token
match comes a long way after last
match. The 3 reads are caused by the
running match, the checksum update and the
literal send. */
if (backup >= s->blength+CHUNK_SIZE && end-offset > CHUNK_SIZE)
matched(f, s, buf, offset - s->blength, -2);
} while (++offset < end);
matched(f, s, buf, len, -1);
map_ptr(buf, len-1, 1);
}
